KAYLA · MODEL 001 · v0.4.0

STILL IDENTITY / SCENE FIRST-FRAME PROMPT
Attach the neutral anchor and nearest face/body angle. Review the scene still before motion.

Show Kayla, fictional adult model 001. Use the attached neutral barefoot anchor and nearest face or body reference. Preserve the established facial proportions, brown almond-shaped eyes, full arched brown brows, softly rounded nose tip, rose-pink lips, warm tan skin, natural texture, chestnut hair, caramel highlights and center part. Preserve shoulder width, torso length, bust proportions, waist-to-hip proportions, limb lengths and natural asymmetry. For body references use the fitted pink unitard with bare feet, no footwear, both feet flat, relaxed neutral stance and arms slightly away from the torso. Height target: provisional fictional 170 cm; never claim this was measured from the image. Scene: {scene}. Wardrobe: {wardrobe}. Expression: {expression}. Camera: {camera}. Lighting: {lighting}. Produce one stable scene first frame. Do not reshape the body beneath the clothes. Hidden geometry is interpreted, not measured.

MOTION-ONLY IMAGE-TO-VIDEO PROMPT
Attach the accepted scene first frame. Pin the exact provider/model/version/settings. Do not rebuild the appearance in this motion prompt.

Animate the attached accepted scene first frame. Action: {one_simple_action}. Camera: {camera_motion_or_locked}. Duration: {seconds}. Keep her existing face, body proportions, outfit, hair, lighting, scene and ground contact stable throughout. Use natural movement and continuous anatomy. Do not add scene changes, new people or wardrobe changes. Dialogue, only if configured: {dialogue_or_none}. Preserve lip sync when audio is enabled. Avoid flicker, face drift, body drift, floating heels, extra fingers and sliding feet.

GATES
Brief -> reviewed scene first frame -> short provider-native test -> identity/anatomy QA -> optional accepted 1080p final -> one justified finishing pass. Initially submit once per shot. Reconcile uncertain paid submissions before retrying. Stop after two motion-test attempts. The four motion tests share a $5 external API cap with no top-ups.

COST AND CACHE
Hash exact references, full brief, prompt, provider/model/version, settings, duration, resolution and audio. Reuse accepted cache entries. Count retries, discarded jobs, audio, LLM, finishing, storage and compute in cost per accepted deliverable. Unknown cost is not zero; self-hosting is not costless. Compare direct and existing aggregators with verified rates, access and quality. Fixed seed is not an identity guarantee. No mass 4K rerolls or promised percentage savings.

DRIFT CHECKS
- Visible face and body match the active barefoot fitted-unitard anchor and nearest reference angle.
- Eye spacing, brows, nose, mouth width, jaw and chin remain stable.
- Hairline, center part, hair color and volume remain stable.
- Skin tone and apparent age remain consistent under intended lighting.
- Body proportions and nominal rotation direction agree with the full-body references.
- Identity remains stable through motion, speech, blinks and the final frame.
- Shoulders, torso, bust, waist, hips and limb proportions stay consistent; clothing does not reshape them.
- Every current reference shows bare feet with natural toes and heel placement; no footwear or leftover straps remain.
- The neutral turn has flat heels and toes on one ground plane, relaxed stance and a clear body silhouette.

COMPLETED REFERENCE QA
The stored speech proof passes calibrated SyncNet and independent ASR for its exact file hash. See qa.json. The active turn reference is the reviewed 85-frame front-to-profile cut; it does not claim the original five-second final hold. Reuse these accepted files. Future scenes still need their own review.